sgl-project/sglang · error · FileNotFoundError

RIFE weight file not found: {flownet_path} Expected layout:

Error message

RIFE weight file not found: {flownet_path}
Expected layout: <model_path>/flownet.pkl

What it means

FileNotFoundError raised by RIFE interpolator load_model when the expected flownet.pkl is not present inside the given model directory. RIFE ships weights as <dir>/flownet.pkl and the loader refuses anything else.

Source

Thrown at python/sglang/multimodal_gen/runtime/postprocess/rife_interpolator.py:288

    def eval(self) -> "Model":
        self.flownet.eval()
        return self

    def device(self) -> torch.device:
        return next(self.flownet.parameters()).device

    def load_model(self, path: str, strip_module_prefix: bool = True) -> None:
        """Load weights from {path}/flownet.pkl.

        Args:
            path: Directory containing ``flownet.pkl``.
            strip_module_prefix: If True, strip the ``module.`` prefix that
                ``DataParallel`` / ``DistributedDataParallel`` adds to keys.
        """
        flownet_path = os.path.join(path, "flownet.pkl")
        if not os.path.isfile(flownet_path):
            raise FileNotFoundError(
                f"RIFE weight file not found: {flownet_path}\n"
                "Expected layout: <model_path>/flownet.pkl"
            )

        def convert(param):
            if strip_module_prefix:
                return {
                    k.replace("module.", ""): v
                    for k, v in param.items()
                    if "module." in k
                }
            else:
                return {k: v for k, v in param.items() if "module." not in k}

        state = torch.load(flownet_path, map_location="cpu", weights_only=False)
        self.flownet.load_state_dict(convert(state), strict=False)
        logger.info("Loaded RIFE weights from %s", flownet_path)

View on GitHub (pinned to 0132848349)

Solutions

1. Ensure the path is a directory containing flownet.pkl exactly
2. Rename your pkl to flownet.pkl if it's the correct RIFE weights
3. Re-download/extract the full RIFE model directory

Example fix

# before
interp.load_model("/models/rife.zip")
# after
# /models/rife/flownet.pkl must exist
interp.load_model("/models/rife")

import os
assert os.path.isfile(os.path.join(model_path, "flownet.pkl")), "missing flownet.pkl"

def rife_model_dir_ok(p: str) -> bool:
    import os; return os.path.isfile(os.path.join(p, "flownet.pkl"))

When it happens

Trigger: Calling load_model (directly or via _ensure_model_loaded before interpolation) with a path that is a zip/gguf, a directory with a differently-named pkl (e.g. 'flownet_v4.pkl'), or an empty/incomplete extraction.

Common situations: Downloading RIFE weights as a raw file instead of preserving directory layout; partial extraction; pointing at the parent of the actual model dir.

Related errors

• Weight file {source.filename!r} was not found in {source.rep
• No model weights found in {path} (expected model.safetensors
• metal shader source not found: {metal_src}
• Prompt text file not found: {path}
• no safetensors files found in {quantized_path or component_m

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/a407ee2bd97c56b1. Report an issue: GitHub.